A Bionic Visual Perception Optimization Method Adapted to Multiple Signal Intensities Within GIS

Bionic visual perception technology captures optical signals from gas insulated metal enclosed switchgear (GIS) partial discharge by mimicking the biological visual system, achieving real-time detection and recognition of partial discharge phenomena under complex electromagnetic environments. However, existing technologies often consider only a single spectrum and do not account for differentiated thresholds for various discharge phenomena, affecting imaging accuracy. This paper proposes a multi-spectral bionic visual perception optimization method for GIS. First, a multi-spectral bionic visual perception framework is constructed. Second, an optimization problem is formulated to maximize the average imaging accuracy of all GIS discharge phenomena. Next, a two-stage event-driven deep Q-network (DQN) optimization method is proposed, learning the optimal light intensity change threshold through two-stage closed-loop feedback, including offline and online learning. Finally, the superior performance of the proposed method is validated through simulations.